Dedicated NVR vs BlueIris with DeepStack?

[Charles_CO]

Hi guys, I have a few not so smart cameras a few years old. I run BlueIris at the moment on pretty weak PC pushing it to the extreme edge.
I want to upgrade to gain some AI capabilities. I am good with IT stuff.
My question is. If you were in my position, would you get dedicated NVR such as Dahua to would you get good PC and use existing BI with DeepStack?

What are the pros and cons? I haven't played with DeepStack yet but it looks like it can do some decent stuff but may not be in it's prime time yet?

Thank, Charles

 

[sebastiantombs]

In terms of flexibility Blue Iris is the choice. If you want it "KISS" then an NVR.

DeepStack works fine with people, cars and some animals. It isn't great if you want facial recognition. In any case it's best to use a dedicated NVidia GPU to handle the DeepStack processing. A GPU works much faster and, to me, more accurately than a CPU can, even a 12th generation CPU. A Quadro with as many cores and as much memory as possible will work, but a 103, 1060 or better will work even better. It all depends on your budget. CUDA capable cards are in the crazy price range right now.

 

[Charles_CO]

It isn't great if you want facial recognition.

Is it because it is not mature enough yet or is it some king of limitation?

In any case it's best to use a dedicated NVidia GPU to handle the DeepStack processing. A GPU works much faster and, to me, more accurately than a CPU can, even a 12th generation CPU. A Quadro with as many cores and as much memory as possible will work, but a 103, 1060 or better will work even better. It all depends on your budget. CUDA capable cards are in the crazy price range right now.

Thanks for the tip. It's good to know. I don't worry too much about money but I want to make good decision.

 

[sebastiantombs]

The problem with DeepStack and facial recognition that I see is that you, basically, need to develop your own model. A model is a collection of captures that get analyzed and stored for use by DeepStack. The more captures for each face in the model, under varying conditions of angles and lighting, the more accurate it can be. That's where the problem comes from. You need to spend hours saving captures for use with DeepStack.

Another thing to consider with facial recognition is that it is normally used with a dedicated camera that always gets a good, full face shot. In reality it is far different than what you see on TV or in the movies.

 

[wittaj]

Are you doing every optimization in the wiki including substreams?

What i# and CPU version to you have

 

[Charles_CO]

Are you doing every optimization in the wiki including substreams?

What i# and CPU version to you have

I know my BI PC is way underpowered. See specs below (attached picture). I have 4 cameras now. 2 of them 4K, recording to NAS with the essential stuff on this PC. I also run Home Automation software on it. So not ideal at all. It works but I know it is pushing the limits and if I want to use DeepStack I will have to upgrade for sure!

 

[sebastiantombs]

Ae you running sub streams on the cameras? If not do so. It will lower CPU utilization by a factor of 50% or so.

 

[wittaj]

What is your CPU%

A member here is running 18 cameras on that CPU at 14% CPU and recently started adding DeepStack to a few of the cameras. Substreams are a must, even more so with a 3rd gen.

A cheap $100ish P400 or GT1030 graphics card could be added to offload DeepStack if it struggles.

 

[Charles_CO]

The problem with DeepStack and facial recognition that I see is that you, basically, need to develop your own model. A model is a collection of captures that get analyzed and stored for use by DeepStack. The more captures for each face in the model, under varying conditions of angles and lighting, the more accurate it can be. That's where the problem comes from. You need to spend hours saving captures for use with DeepStack.

That's not good.. :-(

I like the idea of having more flexibility with BI and DeepStack.

 

[wittaj]

I can assure you that creating your own model for facial recognition for DeepStack will be more successful than the NVR doing it...you cannot train the NVR, but you can DeepStack.

Most of us have found that facial recognition is more gimmicky and novelty than anything else. If you have to put in 5 or 10 or 15 or 35 pictures or more of yourself in the system for it to recognize it is you...then you shouldn't expect much.... My success rate was under 5% so I moved on to other things LOL.

It can work in certain situations like a business that requires everyone to stop in front of the camera and the camera is at head height. Outside of that, the percentage of being accurate is probably not going to be super high. You will get a lot of false "confirmations" doing a search.

Someone here posted once how horrible it was inside his house identifying his neighbors and others as him. Another guy his kids and wife were being tagged as him.

Unless you spend the big bucks that casinos and airports have LOL.

Heck even in ideal situations like a business with the camera at ideal height and optimal lighting it fails....

What do you think of Dahua's "Uyghur Warnings"?

Good luck. Wish you no ill will. I'm not here to argue with people simply to contribute when and where I can. This is also followed by lending my views on current technology and how it impacts all of us. It's undisputed that splitting of the atom changed the world but once again man decided to...

ipcamtalk.com

 

[Charles_CO]

What is your CPU%

A member here is running 18 cameras on that CPU at 14% CPU and recently started adding DeepStack to a few of the cameras. Substreams are a must, even more so with a 3rd gen.

A cheap $100ish P400 or GT1030 graphics card could be added to offload DeepStack if it struggles.

See attached screen shot. Maybe I need to look into optimization regardless of the future upgrade.

 

[wittaj]

See this thread:

My humble i5-4590 still humming along with 30+ cams:

I see people posting about their i7-xxxx system bogging down with less than 10 cameras so I wanted to update my setup. I have ~33 cams running on my BI system: HP Elitedesk 800 G1 tower i5-4590 @ 3.3GHz 16GB ram 12TB total purple drives for storage 30+ total cams currently - 22 cams visible in...

ipcamtalk.com

And the optimizations:

Optimizing Blue Iris's CPU Usage

Blue Iris uses a lot of CPU time. In most Blue Iris installations, your CPU is the limiting factor which determines how many cameras you can have. As such, it is extremely important to optimize your configuration to ensure you can get the most out of your system while consuming the least...

ipcamtalk.com

 

[Charles_CO]

I can assure you that creating your own model for facial recognition for DeepStack will be more successful than the NVR doing it...you cannot train the NVR, but you can DeepStack.

Most of us have found that facial recognition is more gimmicky and novelty than anything else. If you have to put in 5 or 10 or 15 or 35 pictures or more of yourself in the system for it to recognize it is you...then you shouldn't expect much.... My success rate was under 5% so I moved on to other things LOL.

What do you think of Dahua's "Uyghur Warnings"?

Good luck. Wish you no ill will. I'm not here to argue with people simply to contribute when and where I can. This is also followed by lending my views on current technology and how it impacts all of us. It's undisputed that splitting of the atom changed the world but once again man decided to...

ipcamtalk.com

To be honest I don't care too much about face recognition. As long as I can use some AI on my dumb camera to eliminate most of the false motion detections (rain, snow, cloud shadows etc.)

I live in the mountains so do some training for wildlife besides recognizing people and cars would be nice. That is why I am asking about DeepStack.

 

[Charles_CO]

See this thread:

My humble i5-4590 still humming along with 30+ cams:

I see people posting about their i7-xxxx system bogging down with less than 10 cameras so I wanted to update my setup. I have ~33 cams running on my BI system: HP Elitedesk 800 G1 tower i5-4590 @ 3.3GHz 16GB ram 12TB total purple drives for storage 30+ total cams currently - 22 cams visible in...

ipcamtalk.com

And the optimizations:

Optimizing Blue Iris's CPU Usage

Blue Iris uses a lot of CPU time. In most Blue Iris installations, your CPU is the limiting factor which determines how many cameras you can have. As such, it is extremely important to optimize your configuration to ensure you can get the most out of your system while consuming the least...

ipcamtalk.com

I'll check it out..

 

[wittaj]

To be honest I don't care too much about face recognition. As long as I can use some AI on my dumb camera to eliminate most of the false motion detections (rain, snow, cloud shadows etc.)

I live in the mountains so do some training for wildlife besides recognizing people and cars would be nice. That is why I am asking about DeepStack.

Yeah, DeepStack can do that very well if all you care about is a trigger for person or vehicle as an example. It is when you want to to recognize you and unlock your front door and turn off your alarm that it will struggle.

And then yes you could train it for any wildlife that is not already in the model. You would be SOL with an NVR for that.

 

[Flintstone61]

If your not a real busy guy and and have very few motions of interest an NVR could be ok.
If your a real busy guy and have a lot of motion to review. BI will save you shitloads of TIME.

 

[Charles_CO]

Yeah, DeepStack can do that very well if all you care about is a trigger for person or vehicle as an example. It is when you want to to recognize you and unlock your front door and turn off your alarm that it will struggle.

And then yes you could train it for any wildlife that is not already in the model. You would be SOL with an NVR for that.

That was I was thinking. I was leaning towards to BI and DeepStack as I want to get hands on something else than NVR. The best way to learn is to have it physically and experiment.

That said, I wanted to check first with some of you here with experience.

 

[sebastiantombs]

I would say that DeepStack has reduced false alerts by 99% or better for me. We have lots of trees in the yard and trying to "dial in" Blue Iris motion detection was a real chore. DeepStack made it point and click.

 

[Charles_CO]

It is when you want to to recognize you and unlock your front door and turn off your alarm that it will struggle.

I guess for that I could always get good smart camera with alarm output (relay) wires and get it connected alarm system and have that processed with home automation system.. I think there would be way to deal with that.

 

[Charles_CO]

I would say that DeepStack has reduced false alerts by 99% or better for me. We have lots of trees in the yard and trying to "dial in" Blue Iris motion detection was a real chore. DeepStack made it point and click.

Oh yeah! That one of the reasons. Moving clouds is another one! And snow..